microstructure evolution in deformed polycrystals
TRANSCRIPT
HAL Id: hal-02494630https://hal.archives-ouvertes.fr/hal-02494630
Submitted on 28 Feb 2020
HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, estdestinée au dépôt et à la diffusion de documentsscientifiques de niveau recherche, publiés ou non,émanant des établissements d’enseignement et derecherche français ou étrangers, des laboratoirespublics ou privés.
Microstructure evolution in deformed polycrystalspredicted by a diffuse interface Cosserat approach
Anna Ask, Samuel Forest, Benoit Appolaire, Kais Ammar
To cite this version:Anna Ask, Samuel Forest, Benoit Appolaire, Kais Ammar. Microstructure evolution in deformedpolycrystals predicted by a diffuse interface Cosserat approach. Advanced Modeling and Simulationin Engineering Sciences, SpringerOpen, 2020, 7 (1), 10.1186/s40323-020-00146-5. hal-02494630
Ask et al.
RESEARCH
Microstructure evolution in deformed polycrystalspredicted by a diffuse interface Cosserat approachAnna Ask1*, Samuel Forest2, Benoıt Appolaire3 and Kais Ammar2
*Correspondence:
[email protected] Materiaux et
Structures, Onera, BP 72, 92322
Chatillon CEDEX, France
Full list of author information is
available at the end of the article
Abstract
Formulating appropriate simulation models that capture the microstructureevolution at the mesoscale in metals undergoing thermomechanical treatments isa formidable task. In this work, an approach combining higher-order dislocationdensity based crystal plasticity with a phase-field model is used to predictmicrostructure evolution in deformed polycrystals. This approach allows to modelthe heterogeneous reorientation of the crystal lattice due to viscoplasticdeformation and the reorientation due to migrating grain boundaries. The modelis used to study the effect of strain localization in subgrain boundary formationand grain boundary migration due to stored dislocation densities. It isdemonstrated that both these phenomena are inherently captured by the coupledapproach.
Keywords: Cosserat crystal plasticity; phase-field method; diffuse interfaces;dislocation density; grain boundary migration
IntroductionA metallic polycrystal exposed to thermomechanical treatment can undergo sig-
nificant microstructural evolution due to viscoplastic deformation and subsequent
or concurrent recrystallization processes. Heterogeneous reorientation of the crystal
lattice during plastic deformation may for instance lead to fragmentation of existing
grains. In particular, material induced strain localization (slip bands) and curva-
ture localization (kink bands) [1, 2, 3] may be associated with the formation of new
(sub-)grain boundaries [4]. The associated lattice curvature is accommodated by
geometrically necessary dislocations (GNDs). At the same time, statistically stored
dislocations (SSDs) accumulate in the microstructure due to random trapping [5].
The associated stored energy is an important driving force for grain boundary mi-
gration [6, 7, 8, 9], and in the recrystallization process it is observed that dislocation-
free nuclei form which expand into and replace grains of high stored energy content.
Detailed continuum descriptions of grain boundaries can also be based on decompo-
sition into dislocation/disclination distributions [10] or micromechanical concepts
allowing for shear-coupled boundary migration [11]. These latter specific mecha-
nisms are not addressed in the present work.
Viscoplastic deformation at the mesoscale (defined here as the grain scale of
the polycrystal) including the heterogeneous reorientation discussed above can be
successfully modeled by crystal plasticity approaches. In order to create extended
frameworks which also take into account grain boundary migration after deforma-
tion, and in some cases nucleation as an intermediate step, several authors have
Ask et al. Page 2 of 29
coupled crystal plasticity models with methods such as cellular automata [12, 13],
level-sets [14, 15] or phase-field methods [16, 17, 18, 19, 20] (these references are
by no means a complete list but should rather be considered a sample of works
for each mentioned approach). There are only a few works that consider a strong
coupling, formulating and solving the complete set of equations as a monolithic
system [21, 22, 23]. One of the major issues to resolve in a fully coupled framework
is the fact that crystal orientation may change through deformation or through
grain boundary migration. Often, the models that are combined rely on different
descriptors of the microstructure and it is necessary to translate between those in
intermediate processing steps.
Kobayashi, Warren and Carter (KWC) [24, 25] proposed a model that treats
the crystal orientation as a phase-field with diffuse interfaces. Their formulation
couples the evolution of the orientation phase-field to the evolution of a second, more
classical phase-field variable. In this way, the model provides direct access to the
change in crystal orientation that is due to atomic reshuffling processes in the wake
of a migrating grain boundary. The energy density that KWC introduced only takes
the norm of the orientation gradient (i.e., the lattice curvature) as a variable, and
not the orientation itself, in order to respect the requirement of frame invariance.
In particular, the inclusion of a linear term is necessary in order to produce a
solution with regions of constant orientation (grains) separated by localized regions
of non-zero curvature (grain boundaries). It is worth stressing that [26, 27] have
amended this model with a contribution due to statistically stored dislocations, so
as to address stored energy driven grain boundary migration that is the dominant
growth phenomenon during recrystallization.
In this paper, a fully coupled, thermodynamically consistent model that com-
bines a Cosserat single crystal plasticity model with the modified KWC orientation
phase-field model of [26] is used to predict microstructure evolution in deformed
polycrystals. The model was first developed for small deformations [28] and later
for large deformations in [22]. In this work, a linearized version of the large deforma-
tion model (c.f., [21]) is used. The Cosserat continuum is endowed with rotational
degrees of freedom that can be associated with the lattice directors of the crystal.
A strong coupling between the crystal plasticity model and the KWC orientation
phase-field model is made possible via the Cosserat degrees of freedom. The lattice
curvature parameter in the free energy density of the KWC model is replaced in
the coupled model by the curvature or wryness tensor of the Cosserat continuum
[29, 30].
Because the KWC model represents the lattice orientation as a continuous field
rather than relying on a cellular description of the grains, it can accommodate
the heterogeneous reorientation associated with plastic slip processes. Moreover,
because the KWC model introduces a phase field variable, it is able to consider
reorientation by a moving front. In the coupled model adopted in this work, the
grain boundary migration is not accommodated by plastic slip processes, unlike the
model in [23], where the KWC model is combined with strain gradient plasticity.
Indeed, the driving forces for grain boundary migration are capillary forces or stored
energy due to scalar dislocation densities which accumulate or recover following a
modified Kocks-Mecking-Teodosiu evolution law [26, 27, 28].
Ask et al. Page 3 of 29
This paper is organized as follows. The diffuse interface framework is summarized
and the constitutive model is presented. A subsection is dedicated to the identifica-
tion of model parameters. Then follows the numerical examples, with a subsection
dedicated to the formation of (sub-)grain boundaries during deformation and a
subsection dedicated to grain boundary migration. Some first results toward recrys-
tallization are also presented. The paper is then concluded with a summary of the
main results and some points on future work.
Notation
Vectors will be denoted by underline as a and second order tensors by an underscore
tilde symbol as A∼ , with the transpose A∼T . Third order tensors will in the same vein
be denoted by a∼
and fourth order tensors by C≈
. Double contraction is written as
A∼ : B∼ and simple contraction as b = A∼ · a. The scalar product of two vectors can
be written as a ·b = aT b, whereas the dyadic product is written as A∼ = a∼⊗b∼. The
gradient and divergence are written by products using the nabla symbol as a ⊗∇and ∇ · a or A∼ · ∇, respectively, with differentiation of a tensor assumed to act on
the second index. The curl operator is denoted ∇×A∼ .
The second order identity tensor will be denoted I∼. The transformations between
skew-symmetric tensors and pseudo-vectors can be written by means of the Levi-
Civita permutation tensor ε∼
as
×A = −1
2ε∼
: skew(A∼ ) = −1
2ε∼
: A∼ , (1)
and
skew(A∼ ) = −ε∼·×A , (2)
respectively.
The diffuse interface Cosserat frameworkThe small deformation theory presented in this section recalls the formulation pro-
posed in [21], and represents the linearized version of the large deformation theory
developed in [22].
Governing equations
A Cosserat medium is endowed with three additional degrees of freedom at each
material point, allowing for a microrotation which is, a priori, independent of the
displacement degrees of freedom. The additional degrees of freedom necessary to
describe the microrotation can be collected in the pseudo-vector Θ. In the small
deformation setting the Cosserat rotation tensor is then given by
R∼ = I∼ − ε∼ ·Θ . (3)
The linearized, objective deformation measures are given by [29, 30]
e∼ = u⊗∇+ ε∼·Θ , κ∼ = Θ ⊗∇ . (4)
Ask et al. Page 4 of 29
where u is the displacement vector. The curvature or wryness tensor κ∼ = Θ ⊗ ∇can be directly related to the tensor of geometrically necessary dislocation densities
α∼ through a famous result by Nye [31]
α∼ = κ∼T − tr(κ∼) I∼ , κ∼ = α∼
T − 1
2tr(α∼) I∼ . (5)
The derivation of the above expression is detailed in [30].
In the framework of small deformations the Cosserat deformation tensor is de-
composed into elastic and plastic parts as
e∼ = e∼e + e∼
p . (6)
So far, a similar decomposition of the curvature tensor has not been considered in
the coupled theory although this would be possible to introduce as outlined in [32].
In order to allow for microstructure evolution due to grain boundary migration,
[28, 22] coupled the Cosserat model to the orientation phase-field model proposed
by [24, 25]. To this end, a phase-field φ ∈ [0, 1] is introduced as an additional degree
of freedom. The phase-field variable is interpreted in the coupled model as a coarse-
grained measure of crystalline order. In the bulk of an undeformed grain the order
parameter takes the value φ = 1 whereas φ < 1 in grain boundaries. In deformed
grains the build-up of so–called statistically stored dislocations may also result in
the phase-field variable taking values φ < 1. This will be discussed in more detail
when the constitutive model is presented. Note that the crystal orientation, which
was included as a scalar phase-field in the model proposed by [24, 25], is associated
with the Cosserat degrees of freedom in the model presented here.
The derivation of the governing balance equations and boundary conditions re-
lies on the principle of virtual power and is detailed in [28, 22]. The phase-field
contribution is accounted for in the principle of virtual power by the microstress
formalism introduced by [33, 34]. There are therefore in total four work-conjugate
pairs φ : πφ,∇φ : ξφ, e∼ : σ∼ ,κ∼ : m∼ and the principle of virtual power, after some
manipulations, takes the form
∫Dφ[∇ · ξ
φ+ πφ + πext
φ
]dV +
∫∂D
φ[πcφ − ξφ · n
]dS
+
∫Du ·[σ∼ · ∇+ f ext
]dV +
∫∂Du ·[f c − σ∼ · n
]dS
+
∫DΘ ·
[m∼ · ∇+ 2
×σ + cext
]dV +
∫∂DΘ ·
[cc −m∼ · n
]dS = 0 ,
(7)
over any region D of the body Ω with outward unit normal n on the boundary
∂D. Superscripts 〈•〉ext and 〈•〉c denote external body and contact forces and cou-
ples, respectively. It follows that the balance equations and corresponding boundary
Ask et al. Page 5 of 29
conditions for the coupled formulation are given by [28]
∇ · ξφ
+ πφ + πextφ = 0 in Ω , (8)
σ∼ · ∇+ f ext = 0 in Ω , (9)
m∼ · ∇+ 2×σ + cext = 0 in Ω , (10)
ξφ· n = πcφ on ∂Ω , (11)
σ∼ · n = f c on ∂Ω , (12)
m∼ · n = cc on ∂Ω , (13)
where n is the outward normal to the boundary ∂Ω of the body Ω. The first line
(8) gives the balance law for the microstresses related to the phase field φ and its
gradient ∇φ. It should be noted that the stress σ∼ associated with the Cosserat
deformation e∼ is not symmetric, and must therefore not be interpreted as the usual
Cauchy stress. Its skew-symmetric contribution appears also in the balance equation
for the couple-stress m∼ , which in turn is associated with the wryness tensor κ∼. In
this paper, the body forces πextφ ,f ext and cext are assumed to vanish.
Applying the standard thermodynamical principles on a free energy density of
the format ρΨ = ψ(φ,∇φ, e∼e,κ∼, rα), where rα are internal variables related to
the inelastic behavior, results in the following expression of the Clausius-Duhem
inequality [21]:
− ρ Ψ− πφ φ+ ξφ∇φ+ σ∼ : e∼ +m∼ : κ∼ =
−[πφ +
∂ψ
∂φ
]φ+
[ξφ− ∂ψ
∂∇φ
]· ∇φ+
[σ∼ −
∂ψ
∂e∼e
]: e∼
e +
[m∼ −
∂ψ
∂κ∼
]: κ∼
+ σ∼ : e∼p −
∑α
∂ψ
∂rαrα ≥ 0 ,
(14)
assuming isothermal conditions. The constitutive relations are deduced from the
energetic contribution to inequality (14):
πeqφ = −∂ψ∂φ
, (15)
ξφ
=∂ψ
∂∇φ, (16)
σ∼ =∂ψ
∂e∼e, (17)
m∼ =∂ψ
∂κ∼. (18)
Furthermore, a thermodynamic force associated with the internal variables rα can
be introduced as
Rα =∂ψ
∂rα. (19)
In order to recover the phase-field dynamics for the variable φ, it is assumed in line
with [34, 35] that the stress πφ contains stored and dissipative contributions so that
Ask et al. Page 6 of 29
πφ = πeqφ + πneqφ , (20)
with only the stored part given by (15).
Evolution equations governing the dissipative processes are assumed to be deriv-
able from a potential (with the appropriate convexity properties to ensure non-
negative dissipation)
Ω = Ωp(σ∼) +Ωα(Rα) +Ωφ(πneqφ ) , (21)
so that
e∼p =
∂Ωp
∂σ∼, rα = −∂Ω
α
∂Rα, φ = − ∂Ωφ
∂πneqφ
. (22)
Link between the lattice rotation and the Cosserat microrotation
The rate of the Cosserat deformation can be written as
e∼ = ε∼ + ω∼ + ε∼· Θ , (23)
where the symmetric contribution is the rate of the usual small strain tensor ε∼ =12 [u⊗∇+∇⊗ u ], and ω∼ = 1
2 [ u⊗∇−∇⊗ u ] is the spin tensor. The skew-
symmetric part of the above expression can be collected in the pseudo-vector
×e =
×ω − Θ . (24)
Equation (24) gives the relation between the relative rotation of the material and
the Cosserat microrotation of a triad of microstructural directors attached to the
material point.
By defining the elastic and plastic spin pseudo-vectors×ω p :=
×e p and
×ω e :=
×ω − ×ω p, respectively, the rate of the skew-symmetric deformation (24) can now be
expressed as
×ω e − Θ =
×e e . (25)
The above equation provides the necessary link between the rotation of the crystal
lattice vectors and the Cosserat directors that is an essential part of the modeling
framework. They can be made to remain parallel by enforcing the internal constraint
×e e ≡ 0 . (26)
The Cosserat crystal plasticity framework thereby allows for a direct access to
the heterogeneous evolution of the crystal orientation in the polycrystal during
deformation.
Ask et al. Page 7 of 29
Constitutive model and identification of parametersIn this section the specific constitutive choices are presented and the identification
of model parameters is described. Pure copper is used as a model material. The
resulting model is a non-local Cosserat crystal plasticity model with diffuse and
mobile grain boundaries. Grain boundary migration is either driven by the interface
curvature or by the energy stored during plastic deformation.
Constitutive model
The free energy density for the coupled problem is chosen as [28]
ψ(φ,∇φ, e∼e,κ∼, r
α) = f0
[f(φ) +
a2
2|∇φ|2 + s g(φ)||κ∼||+
ε2
2h(φ)||κ∼||
2
]+
1
2ε∼e : E
≈s : ε∼
e + 2µc×e e · ×e e + ψρ(φ, r
α) ,
(27)
with
ψρ(φ, rα) = φ
N∑α=1
1
2µ rα2 , (28)
where N is the number of crystal slip systems and rα are internal variables. This
contribution is taken to represent stored energy due to random trapping of dislo-
cations during plastic deformation by associating the internal variables with the
statistically stored dislocation densities, here denoted ρα, according to
rα = b
√√√√ N∑β=1
hαβρβ . (29)
Inserted into equation (28), this then gives the following expression for the energy
contribution related to the scalar dislocation densities ρα:
ψρ(φ, rα) =
1
2µ b2 φ
N∑α=1
N∑β=1
hαβρβ . (30)
This stored energy is an important driving force for grain boundary migration [7, 8].
The coupling with the phase-field variable by means of the linear term in φ ensures
that local gradients in the stored energy result in grain boundary migration [26, 27].
In the grain boundary regions, φ tends toward zero. This means that energy is
stored mainly in the bulk, and not inside the grain boundaries where the concept
of statistically stored dislocations loses its significance.
Whereas the contribution due to the phase-field φ in the energy density (27) con-
tains standard bulk and interphase terms, the contribution due to the curvature,
which can be considered as a generalization to three dimensions of the energy contri-
bution proposed by [24, 25] for the orientation phase-field, must contain the rank one
term ||∇κ∼||. It is this term that ensures a solution with localized grain boundaries.
At zero lattice curvature this term renders the energy density non differentiable
and in the numerical treatment a regularization is therefore applied [36, 28]. In the
Ask et al. Page 8 of 29
context of non-local crystal plasticity, several authors have considered energy den-
sities that contain linear contributions due to the GND density tensor [37, 38, 39],
interpreting the linear term to represent the self energy of the GNDs whereas the
quadratic term represents the energy due to their interactions. It may be noted
that [40] found solutions with substructure partitioning, reminiscent of subgrains,
even without a rank one term for a large deformation Cosserat crystal plasticity
model. While those results are not directly applicable to the coupled model, it may
nevertheless be warranted to reconsider the optimal format of the energy density in
the future.
The functions g(φ) and h(φ) are required to be non-negative. A simple quadratic
function is chosen for the latter such that h(φ) = φ2 whereas, following [24], a
logarithmic function is chosen for g(φ) such that
g(φ) = −2 [ log(1− φ) + φ ] . (31)
This choice allows to fit a Read-Shockley type behavior for low angle grain bound-
aries. Due to the singularity for φ = 1, a small positive constant is added in the
logarithm in the numerical treatment.
In the second line in (27), the symmetric elastic deformation is included via a stan-
dard quadratic form, where E≈s is the usual elasticity tensor. The skew-symmetric
elastic deformation is penalized by choosing the Cosserat parameter µc sufficiently
large and thereby approaching the constraint (26) that the Cosserat microrotation
should follow the lattice reorientation [30, 41, 42, 43, 40]. This constraint could
alternatively be directly imposed by using Lagrange multipliers.
The stress, couple stress and microstresses take the following forms:
sym(σ∼) =E≈s : ε∼
e , (32)
×σ = 2µc
×ee (33)
m∼ = f0
[s g(φ)
1
||κ∼||+ ε2 h(φ)
]κ∼ , (34)
πeqφ = − f0[
1− φ− s ∂g∂φ||κ∼|| −
ε2
2
∂h
∂φ||κ∼||
2
]−
N∑α=1
1
2µ rα2 , (35)
ξφ
= f0 a2∇φ . (36)
From (33) together with (24), it is apparent that a migrating grain boundary,
which locally reorients the lattice as a result of atomic reshuffling processes, would
give rise to a skew-symmetric stress. In order to relax this stress and allow the
grain boundaries to migrate (without storing elastic energy), the plastic dissipation
potential Ωp is assumed to consist of two terms:
Ωp =
N∑α=1
Kv
n+ 1
⟨|τα| −Rα/φ
Kv
⟩n+1
+1
2τ−1? (φ,∇φ,κ∼)
×σ · ×σ , (37)
where 〈•〉 = Max(•, 0) and×σ contains the skew-symmetric contributions to the
stress. The choice (28) results in a Taylor type hardening law for the associated
Ask et al. Page 9 of 29
force Rα
Rα =∂ψ
∂rα= φ µ rα = φ µ b
√√√√ N∑β=1
hαβρβ , (38)
which is interpreted as the critical resolved shear stress on slip system α. Equation
(37) has been constructed so that the value of the phase-field φ will not influence
the plastic slip flow evolution.
The first term of (37) leads to a classic crystal flow rule. The resolved shear stress
τα on slip system α can be calculated as
τα = `α · σ∼ · nα , (39)
where `α and nα are respectively the slip direction and normal to the slip plane.
Note that the additional contribution to the resolved shear stress due to the non-
symmetry of the stress tensor can be interpreted as a size-dependent kinematic
hardening [44, 4, 45, 46]. The evolution of plastic deformation follows from (22)
and (37)
e∼p =
N∑α=1
γα `α ⊗ nα + τ−1? (φ,∇φ,κ∼) skew(σ∼ ) , (40)
with the slip rate given by
γα =
⟨|τα| −Rα/φ
Kv
⟩nsign τα . (41)
The second term in (40) allows the relaxation of the skew-symmetric stress by acting
on the plastic spin ω∼p. The viscosity type term τ?(φ,∇φ,κ∼) is constructed so that it
is small in the interface region but large in the bulk of the grains, thereby restricting
the relaxation to the grain boundaries.
There are therefore two contributions to the plastic spin
×ω? = τ−1? (φ,∇φ,κ∼)
×σ , (42)
ω∼p − ω∼
? = skew(
N∑α=1
γα `α ⊗ nα ) , (43)
due to the relaxation of the skew-symmetric stress at moving interfaces and to plas-
tic slip processes, respectively. The skew-symmetric plastic deformation, likewise,
can be considered to consist of two separate contributions
×ep =
×eslip +
×e? , (44)
with
×eslip =
×ωp − ×ω? , ×
e? =×ω? . (45)
Ask et al. Page 10 of 29
The elastic skew-symmetric deformation is then given by
×ee =
×ϑ− ×eslip − ×e? −Θ , (46)
where×ϑ is the pseudo-vector of the skew-symmetric tensor ϑ∼ = skew(u ⊗ ∇ )
denoting the material rotation.
In the present model, the Cosserat microrotation is associated with the lattice
(crystal) orientation as measured with respect to some fixed frame. It is therefore
in general non-zero in the reference configuration. It follows [28, 21] that the ini-
tial plastic deformation must be non-zero for the reference configuration to remain
stress-free. Specifically, this is obtained by adopting the initial condition
×ep(t = 0) =
×e?(t = 0) = −Θ(t = 0) . (47)
The part of the inelastic deformation that is not due to slip processes during de-
formation can therefore be seen as necessary to accommodate an initial orienta-
tion distribution created from a previous, homogeneous state. It can be interpreted
therefore, somewhat simplified, as representing a reference orientation state of the
grain, and its associated evolution law ensures that it is inherited to a region swept
by a migrating grain boundary. The evolution of the plastic deformation due to
migrating grain boundaries according to equation (45) was reported in [22].
The standard evolution equation is obtained for the phase-field φ by choosing a
quadratic dissipation potential [33, 26]
Ωφ =1
2τ−1φ
(πneqφ
)2 , (48)
where τφ is a viscosity type parameter, so that
τφ φ = −πneqφ . (49)
Instead of formulating a dissipation potential for the internal variables, which
are associated with the scalar dislocation density in equation (29), modified Kocks-
Mecking-Teodosiu evolution laws are used, given by
ρα =
1b
(1K
√∑β ρ
β − 2dρα)|γα| − ρα CD A(|κ∼|) φ if φ > 0
1b
(1K
√∑β ρ
β − 2dρα)|γα| if φ ≤ 0
(50)
The additional recovery term, with parameter CD, was originally introduced by
[26, 27]. It allows for static recovery behind the front of a migrating grain boundary,
with the amount of recovery governed by the parameter CD. The static recovery is
localized to the interface region by the function A(|κ∼|) = tanh(C2A ||κ∼||2). The com-
peting nucleation and annihilation mechanisms in equation (50) determine whether
energy is stored or dissipated. In particular, the additional recovery term allows for
stored energy to be released through grain boundary migration.
Ask et al. Page 11 of 29
Table 1 Model parameters used in the simulations.
Parameter value unit commentElasto-viscoplastic parameters
C11 160 GPa Elasticity moduli for cubic symmetryC12 110 GPaC44 75 GPaµc/f0 11.5 GPa Cosserat modulusµ 46 GPa Mean shear modulusb 0.2556 nm Burgers vector (magnitude)Kv 10 MPa s1/n Viscoplasticity parametersn 10K 1/10 Kocks-Mecking parametersd 10 nmCD 100 Determines the rate of static recoveryCA 1 µm
Phase-field and mobility parametersf0 1.15 MPa Sets the magnitude of the grain boundary energys 1.5 µma 0.62 µmε 2 µm Sets the grain boundary width
τφ/f0 100 s Viscosity type (mobility) parameter for φ
τ?/f0 10 s Viscosity type parameter for×ω?
Identification of model parameters
The model parameters used in the simulations are given in table 1. While typical
values for most of the elasto-viscoplastic material parameters of pure copper at
room temperature can be found in the literature (in this paper values adapted from
[47, 48] are used), table 1 also gives values for the recovery parameter CD in (50)
and the Cosserat coupling modulus. The former is chosen so that full static recovery
takes place in the wake of migrating grain boundaries. For the moment, only self
interaction between static dislocations is considered. In equation (29), hαβ = 0.3 if
α = β and otherwise hαβ = 0 if α 6= β. The Cosserat modulus is chosen sufficiently
large to act as a numerical penalty parameter.
The parameters f0, a, and s, related to the phase-field inspired energy contribution
in (27), are calibrated to fit values obtained from atomistic calculations for < 100 >
tilt grain boundaries. Following [28], the general calibration procedure relies on the
matched asymptotic analysis of the original KWC model as proposed in [49]. At the
lowest order corresponding to equilibrium, the coupled model becomes equivalent
to the KWC model. Hence, using the results from [49], it is possible to calculate
the grain boundary energy as a function of the misorientation angle. Some grain
boundary energies – calculated by atomistic simulations [50, 51, 52] for copper
tilt grain boundaries – are collected in figure 1 (top). The grain boundary energy
exhibits cusps, i.e., favored misorientations which result in local energy minima. The
current model does not take into consideration such local minima so the calibration
was only performed on the first part of the curve up to a misorientation angle of 30.
The result is shown with stars in figure 1 (bottom) and the calibrated parameters
are given in table 1. The parameter ε controls (together with a) the width of the
grain boundary. It is chosen to be ε = 2 µm, which results in a diffuse interface
width of around 500 nm. The interface width puts a restriction on the element size
in the finite element discretization. At least four to five nodes, but ideally more, are
required to resolve the curvature.
Whereas the coupled Cosserat model and the model proposed in [26] (which is
close to the original KWC model) become similar at equilibrium, they differ in dy-
Ask et al. Page 12 of 29
Figure 1 Top: Grain boundary energy as a function of misorientation for copper, obtained fromatomistic calculations [50, 51, 52]. Bottom: Result from calibration of model parameters for theorientation phase-field model (black stars) to fit the calculated grain boundary energy of a< 100 > tilt grain boundary from [50] (solid blue line). Only angles up to 30 are considered inthe calibration because the model does not capture the cusps seen in the full range ofmisorientation angles.
namic situations. The two formulations have been compared in [28, 21]. The model
in [26] relies on a relaxation dynamics for the orientation, whereas in the coupled
model the orientation is at each instant governed by the constraint (46) together
with the balance equation (34). Because of this difference, it is not possible to use
the asymptotic analysis of the KWC model to calibrate the mobility parameters of
the coupled model, because only the expansion to lowest order is valid in this case.
The choice of the parameter τ?(φ,∇φ,κ∼) in equation (42) was studied in [28]
where it was shown that it must be bounded by the choice of τφ in order not to
restrict the grain boundary mobility. It will be assumed to be of the general form
τ?(φ,∇φ,κ∼) = τ? P?(φ, ||κ∼||) . (51)
The function P?(φ, ||κ∼||), which is meant to localize the stress relaxation to the
interface region, will be discussed in the next section. In concordance with [28], the
parameter τ? is chosen to be τφ/10. Both values are given (scaled by f0 for clarity)
in table 1. The choice of the parameters for the dynamic behavior is revisited in the
next section when the numerical simulations are discussed. A formal procedure for
identifying the dynamic coefficients of the model remains to be established, which
Ask et al. Page 13 of 29
is why in this work the mobility parameter τφ is adjusted by running numerical
simulations on simplified test cases.
Microstructure evolution in deformed polycrystalsIn this section, the microstructure evolution predicted by the proposed model in
deformed polycrystals is investigated by numerical simulations. In particular, the
effect of localized reorientation of the lattice and the grain boundary migration due
to statistically stored dislocation densities are studied. The numerical simulations
are performed using the finite element code Z–set [53]. The discretization of the weak
form of the balance equations results in a monolithic, coupled system of equations
which is solved using implicit integration in an iterative Newton-Raphson scheme.
The constitutive equations are integrated using a fourth order Runge-Kutta method
with automatic time-stepping. For details on the numerical implementation, see
[28, 22] and also [41] and [26] for the Cosserat crystal plasticity and the phase-field
model, respectively. Since only planar problems are considered, Θ = [ 0 0 θ ]T ,
with x3 considered to be the out-of-plane axis and θ3 = θ the only non-zero ori-
entation degree of freedom. The planar problem therefore has only four degrees of
freedom: two components of displacement, the microrotation θ, associated with the
crystal orientation, and the phase-field φ.
Periodic geometries and boundary conditions are considered. Simple shear loading
is then applied by imposing
u = B∼ · x+ p (52)
with the tensor B∼ given by
B∼ =
0 B12 0
B21 0 0
0 0 0
, (53)
with only either B12 or B21 non-zero. The vector p is a periodic fluctuation that
takes the same value in opposite points of the boundary.
Grain boundary mobility
As discussed in the previous section, the grain boundary mobility cannot be cali-
brated based on results for the orientation phase-field model, as the evolution equa-
tions for the two models are different. While a formal calibration procedure remains
to be established, the grain boundary mobility can be estimated and adjusted by
running a simple test case. This test case is a periodic bigrain of the type considered
in [28], with a flat grain boundary and 15 misorientation between the two grains.
One grain is assigned a (constant) scalar dislocation density and the other grain
is assumed to be dislocation-free. The resulting stored energy gradient acts as a
driving force for grain boundary migration. With a planar grain boundary and a
constant stored energy gradient, the grain boundary mobility can be calculated as
M = v/ψρ, with v being the grain boundary velocity and ψρ the pressure acting
on the grain boundary due to the stored energy.
Ask et al. Page 14 of 29
In the viscosity type parameter τ?(φ,∇φ,κ∼) in equation (51), the localizing func-
tion is chosen to be
P?(||κ∼||) = 1−[ µpε− 1
]exp(−βp ε ||κ∼||) , (54)
following [25]. In [28], the same localizing function was used for the orientation
phase-field, whereas a tanh function was used for the τ? coefficient. The constant
µp determines the value of P?(||κ∼||) in the bulk of the grains and should be chosen
to be large so that 1/τ? → 0 there. The constant βp sets the width of the region
around the grain boundary where P?(||κ∼||) ≈ 1 and should be chosen βp ≤ 103 in
order to achieve localization. In the simulations, µp = 109 µm and βp = 102 are
used.
With the driving pressure due to the stored dislocation density given by ψρ = 0.23
MPa and using the material parameters given in table 1, the resulting simulated
grain boundary velocity is v = 0.8 nm/s. The resulting mobility is M = 3.5
nm/(MPa s). While there is not a plethora of available experimental data on the
grain boundary mobility, [54] measured an average grain boundary mobility of 0.61
nm/(MPa s) at 121C for recrystallization of cold deformed copper. Theoretical
average grain boundary mobilities were also calculated in [54]. Compared to that
data,M = 3.5 nm/(MPa s) corresponds to recrystallization at a temperature of ap-
proximately 150-160C. In the present work, thermal effects are not accounted for
in the simulations, and the indicated temperature range should not be considered
predictive of the true final temperature of the system.
In order to compare the behavior of the coupled model and the orientation phase-
field model, the latter was also used to run the same example of a periodic bi-
grain with a stored energy gradient. The model parameters, where applicable, were
the same as for the coupled model, and in the evolution equation for the orienta-
tion phase-field (c.f. [24, 26, 28]), the mobility parameter was chosen identical to
τ?(φ,∇φ,κ∼). The mobility predicted by that simulation wasM = 0.87 nm/(MPa s),
which is lower than for the coupled model. This is as expected and further indicates
that it is necessary to establish a separate calibration procedure for the mobility of
the coupled model.
Initial conditions for polycrystal simulations
In general, initial microstructures for polycrystalline aggregates are obtained from
measurements of real samples, or generated by tessellation using dedicated software
such as Neper [55]. The tessellations can then be translated into a finite element
mesh with discrete regions of different orientation, i.e., grains. In this work, two
periodic, planar microstructures generated with Neper are used, one with six grains
and one with 32 grains. The smaller geometry is 20 × 20 µm in size, discretized
with 8464 (92 × 92) elements with quadratic interpolation functions and reduced
integration. The larger geometry is 50 × 50 µm, and it is discretized with 47089
(217×217) elements. The element size is therefore nearly the same in both examples,
0.21 µm in the six grains geometry compared to 0.23 µm in the one with 32
grains. The orientations are distributed in the range 0 to 35, which corresponds
to the range of misorientations in the grain boundary energy for which the model
is calibrated.
Ask et al. Page 15 of 29
Figure 2 Initial orientation distribution generated with Neper (left) and the orientationdistribution after initial simulations with the orientation phase-field model (right). The orientationis given in radians. Because the orientation behaves as a phase-field, the grain boundaries in thecoupled model are not restricted to follow the finite element grid and so there is no need to adaptthe mesh to the morphology of the microstructure.
The grain morphologies and orientation distributions obtained using Neper are
not suitable for use as initial conditions in the coupled model as they are far from
the equilibrium solution, with jagged edges tied to the finite element grid, and so
will result in convergence problems (figure 2, left). In order to generate suitable
initial conditions for the non-zero component θ of×Θ as well as φ, simulations using
the KWC orientation phase-field model are performed. This will also generate initial
conditions for the non-zero component×ep of
×ep (c.f., equation (47)). The simulations
take the grain morphologies generated by Neper as initial conditions, together with
a uniform φ = 1. The simulations are run until a solution with localized, smooth
grain boundaries is found. Figure 2 shows the microstructure generated by Neper
for the geometry with six grains and the corresponding output for the orientation
field from the phase-field simulation. This field will serve as initial conditions on θ
and -×ep in the coupled simulations, as shown in figure 3 (top left). The phase-field
simulations also find the corresponding solution for φ, shown in figure 3 (bottom
left). Initial conditions for the geometry with 32 grains are generated by the same
procedure. In all simulations, the dislocation densities have a uniform initial value
of ρα = 2 · 1011 m−2.
Kink bands and subgrain formation
Up to four possible slip systems are considered for the planar problem with the slip
directions given by
l1 = ( 1, 0 ) , l2 = ( 0, 1 )
l3 =1√2
( 1, 1 ) , l4 =1√2
(−1, 1 ) .(55)
The slip plane normal in each case is perpendicular to the slip direction. By choosing
which slip systems are allowed to be active in a simulation, localization phenomena
such as kink or slip bands can be triggered. In particular, the formation of kink
bands, which are bands of localized plastic deformation that are perpendicular to
the slip direction, can result in large localized reorientation of the lattice (see the
recent contribution [3] for a detailed description of kink and slip bands in crystals
Ask et al. Page 16 of 29
Figure 3 The six grain microstructure before and after applied shear deformation. The initialvalues are shown to the left for θ (top), φ (middle) and norm of the curvature |∇θ| (bottom). Topright shows the orientation distribution after shear deformation, middle right shows thecorresponding phase-field φ and bottom right shows the norm of the curvature. The truemaximum value of the curvature norm is around |∇θ| = 1.9 µm−1 but it has been capped off at1 µm−1 for a better visualization. In the simulation, only one slip system α = 2 was active,resulting in the formation of kink bands and subgrains. The slip direction of each grain isindicated with black lines in the top right figure.
and how they arise in crystal plasticity simulations). In the simulations, a periodic
shear deformation is applied with B21 = 0.05 in equation (52). Kink bands are
triggered when only one slip system can be activated.
Figure 3 shows the orientation field θ (top), the phase-field φ (middle) and the
curvature norm |∇θ| (bottom) before (left) and after (right) deformation of the six
grains microstructure. The deformation is not shown, i.e., the fields are projected
onto the undeformed mesh (the total shear deformation is five percent). Only slip
system α = 2 was allowed to activate in the simulation. The formation of kink
bands is most easily seen in the bottom right figure, which shows the norm of the
curvature |∇θ| in the deformed structure. These bands are also visible in the phase-
field φ. Note that the grains are oriented between 0 − 35 from the x1-direction,
so the localized bands are perpendicular to the slip direction since it is the (0, 1)
slip direction that is active. In the top left image, the slip direction of each grain
Ask et al. Page 17 of 29
Figure 4 The six grain microstructure before and after applied shear deformation. The initialvalues are shown to the left for θ (top), φ (middle) and norm of the curvature |∇θ| (bottom). Topright shows the orientation distribution after shear deformation, middle right shows thecorresponding phase-field φ and bottom right shows the norm of the curvature. The truemaximum value of the curvature norm is around |∇θ| = 1.9 µm−1 but it has been capped off at1 µm−1 for a better visualization. In the simulation, only one slip system α = 1 was active,resulting in the formation of kink bands. The localization is less pronounced than in the casewhere only slip system α = 2 was active. The slip direction of each grain is indicated with blacklines in the top right figure.
is indicated with black lines. The orientation phase-field method always localizes
grain boundaries where there is a sufficient orientation gradient [24], and clearly
this property is inherited to the coupled model, as may be expected. In particular,
a kink band has formed across the bottom part of the left central (green) grain and
the right central (blue) grains, resulting in what appears to be two, maybe three,
new grain boundaries delimiting new subgrains. The locations of the new grain
boundaries are near the spots marked 1 and 2, respectively, in the top right image.
Here, the term subgrain is used to denote regions that result from partitioning of the
original grain structure due to the deformation. The model automatically handles
localization and, while it is sensitive to the magnitude of the orientation gradient,
does not differentiate between different types of grain boundaries.
Figure 4 shows the results from calculations when only slip system α = 1 was
allowed to activate. Kink band formation is again observed but the localization is
Ask et al. Page 18 of 29
Figure 5 The 32 grain microstructure before and after applied shear deformation. The initialvalues are shown to the left for θ (top), φ (middle) and curvature |∇θ| (bottom). Top right showsthe orientation distribution after shear deformation, middle right shows the correspondingphase-field φ and bottom right shows the curvature. The true maximum value of the curvaturenorm is around |∇θ| = 1.7 µm−1 but it has been capped off at 1 µm−1 for a better visualization.In the simulation, only one slip system was active, resulting in the formation of kink bands andsubgrains.
less pronounced than in the previous example. Slip band formation is not observed.
The figure shows the orientation field θ (top), the phase-field φ (middle) and the
curvature norm |∇θ| (bottom) before (left) and after (right) deformation.
Figure 5 shows the same results as figure 3, but for the larger geometry with
32 grains. Again, the formation of kink bands is most notable in the contour map
showing the norm of the orientation gradient |∇θ| in the deformed microstructure
(bottom right), but it is also visible in the phase-field φ (middle right). In particular,
below the spot marked with 1 in the top right image, a little bit above the center
of the geometry, there are two new subgrains that seem to have formed due to
the reorientation at the edges of two existing grains. The simulation shows how a
kink band associated with a strong lattice curvature transforms into a true grain
boundary, which is associated with a significant decrease in φ due to the large
accumulation of dislocation densities.
Ask et al. Page 19 of 29
Figure 6 The accumulated plastic slip γ (left) and the scalar dislocation density ρ (right) withonly one active slip system. The maximum display value for ρ has been set to 5 · 1014 m−2 (thescale in the colorbar is in µm−2) in order to more clearly demonstrate how it is distributed overthe grains. The true maximum value is 2.4 · 1015 m−2. For the same reason, the maximum displayvalue for γ has also been limited (at 0.3). In a very narrow region larger values are observed, witha maximum of almost 1.
The interest of these examples is that by triggering kink deformation in the sim-
ulations, it is possible to study how the model behaves when very localized reorien-
tation takes place. Clearly, the coupled framework allows for the formation of new
subgrains by reorientation without the need for post-processing or reinterpretation
of the results.
Dislocation driven grain boundary migration
During the elasto-viscoplastic deformation, energy is stored in the bulk of the grains
due to random trapping of dislocations. This is included in the model via the energy
term (28) and the modified Kocks-Mecking-Teodosiu evolution law (50) for the evo-
lution of the scalar dislocation densities ρα. Because the energy contribution (28)
is coupled with φ in the energy density, a difference in stored energy between two
grains contributes to grain boundary migration. This effect is illustrated by com-
paring simulations performed with only one active slip system (the same as in the
previous example) or all four slip systems of (55). In the former case, the difference
in stored energy is higher, and correspondingly there is a stronger contribution to
the microstructure evolution due to the statistically stored dislocation densities.
Figure 6 shows the accumulated plastic strain (left) and static dislocation density
(right) in the case of only one active slip system for the six grain geometry. The
maximum scalar dislocation density is in reality 2.4 ·1015 m−2, but in the figure the
maximum display value has been limited to 5 ·1014 m−2 in order to better highlight
how the dislocation density is distributed between different grains. In figure 7, the
scaled stored energy ψρ/f0 with one (left) or four (right) active slip systems is
shown. When all four slip systems are active, energy is stored much more evenly in
the grains. The maximum value is around 0.9 for one possible slip system, whereas it
is only around 0.1 with four possible slip systems, so its contribution to the driving
force for grain boundary migration in the latter case is considerably smaller. Note
that in figure 7, the maximum display value in the left figure (one possible slip
system) has been limited to 0.5 in order to better represent how the stored energy
is distributed in the different grains.
The microstructure evolution is studied by applying a shear deformation and then
holding it constant for two hours under a moderately elevated temperature. The
Ask et al. Page 20 of 29
Figure 7 The non-dimensionalized stored energy ψρ/f0 with one (left) or four (right) active slipsystems. The scale for the left-hand image has been cut-off at 0.5 in order to more clearly showhow the stored energy is distributed in the different grains. The true maximum value is aroundψρ/f0 = 0.9. The scale for the right-hand side image has not been altered.
grain boundary mobility was chosen to correspond with a (constant) temperature
of around 150C. This value is not predictive of the true temperature of the system;
it would be necessary to fully take into account thermal effects to achieve this. The
final microstructure for the six grains microstructure is shown in figure 8 for one
(left) or four (right) possible active slip systems. The images show, from top to
bottom, the orientation field, the phase-field φ and the curvature |∇θ|, respectively.
The initial configuration is the same for both cases and the initial grain boundaries
are outlined in black.
It is clear that the predicted grain structures are very different. With only one
active slip system, it seems that the grain boundary migration is primarily driven
by the difference in stored energy associated with statistically dislocation densities.
With four possible slip systems on the other hand, it appears that the grain bound-
ary migration is driven mostly by the grain boundary curvature. This is evident
from the shrinking of the smallest grain, which has a slightly lower stored energy
than its neighbor to the left and therefore should expand in that direction if this
had been the main driving force. Instead, it shrinks to lower the energy in the triple
junction. It is also clear from comparing the two examples that the stored energy
due to the scalar dislocation densities is a stronger driving force than curvature for
grain boundary migration in this case. This is in agreement with the typical driv-
ing pressures given in [7], with typical driving pressures due to stored dislocations
generally several magnitudes larger than those due to the grain boundary energy.
To study the evolution of the statistically stored dislocations, the simulation with
one possible active slip system is considered again. In figure 9, the dislocation density
in the grains is compared at two time points: just after the deformation is applied
(left) and after two hours (right). The grain with the highest dislocation density
post deformation has vanished completely after two hours, and in the wake of the
migrating grain boundaries, static recovery has left a region free of statistically
stored dislocations.
The larger geometry with 32 grains is also studied. Figure 10 shows, from left to
right, the orientation field, the phase-field φ and the norm of the curvature |∇θ| after
one hour of final deformation maintained constant. The initial conditions and the
corresponding fields immediately after the deformation is applied are given in figure
Ask et al. Page 21 of 29
Figure 8 The microstructure after two hours of constant shear deformation at moderatelyelevated temperature. The simulations allow one (left) or four (right) possible active slip systems.From top to bottom the images show the orientation field, the phase-field φ and the curvature|∇θ|. The original grain boundaries are outlined in black. The initial microstructure is the same inboth cases.
Figure 9 The scalar dislocation density ρ with only one active slip systems just after deformationhas been applied (left) and two hours later (right). The maximum display value for ρ has been setto 5 · 1014 m−2 (the scale of the colorbar is in µm−2) in order to more clearly demonstrate how itis distributed over the grains. The true maximum value is 2.4 · 1015 m−2.
5. The microstructure has evolved relatively little in the time interval considered.
In part, this may be due to the distribution of the stored energy inside the grains,
Ask et al. Page 22 of 29
Figure 10 The microstructure of the 32 grain geometry after holding the deformation constantfor one hour. Top left: the orientation field, and top right: the phase-field φ, and bottom: thenorm of the curvature |∇θ|. There is only one possible slip-system.
Figure 11 The dislocation density distribution in the 32 grain microstructure immediately afterdeformation (left) and after 35 minutes of constant deformation (right). The maximum displayvalue for ρ has been set to 5 · 1014 m−2 (the scale of the colorbar is in µm−2) in order to moreclearly demonstrate how it is distributed over the grains. The true maximum value is 1.9 · 1015m−2. There is only one possible slip system in the simulations.
which may be less favorable to grain boundary migration than in the smaller sample.
The maximum value of the stored energy is also lower than in the previous example,
approximately 1.9·1015 m−2 (in figure 11, the maximum display value for ρ has been
set to 5 · 1014 m−2). In the smaller sample, most of the microstructure evolution
takes place within the first hour, so the shorter time span considered in the larger
example is not in itself sufficient to explain the difference.
Observed phenomena for further investigation
For both considered geometries, there are some phenomena that appear in the
simulations that may require further investigation. Figure 12 shows a zoom on the
subgrain boundary that has formed in the lower left corner of the six grain geometry
Ask et al. Page 23 of 29
Figure 12 A zoom on a new (sub-)grain boundary generated in the six grain microstructure dueto kink band formation, c.f. figure 3 and 6 (the grain boundary appears in the lower left corner ofthe microstructure in those figures). From top to bottom the images show the orientation, thedislocation density ρ, the norm of the curvature |∇θ|, the phase-field φ, and the resolved shearstress τ (ranging from -160 MPa to 230 MPa). The evolution is shown from left to right: justafter deformation and then after five minutes, one hour, and two hours of deformation. It appearsthat the reoriented bands (blue above and yellow below) are slightly fragmented, with scattereddarker spots. As the microstructure evolves, the reoriented bands start to vanish except for thesespots, which appear to remain. In the stored energy, blue regions appear at the same positions,indicating that static recovery has taken place there.
after deformation, c.f. figure 3 and 6. Only one slip system is allowed. From top to
bottom the images show the orientation, the dislocation density ρ, the norm of the
curvature |∇θ|, the phase-field φ, and the resolved shear stress τ (ranging from
-160 MPa to 230 MPa). The time evolution is shown from left to right: just after
deformation and then after five minutes, one hour, and two hours of deformation.
Already immediately after the deformation is applied, it seems that the reoriented
bands (blue above and yellow below) are not uniform and that there are small spots
of darker color dispersed inside them. Already after five minutes, and even more so
after one hour, the reoriented bands have mostly vanished (due to local reorienta-
tion) except for these spots. Furthermore, in the same position, there appears in
the scalar dislocation density small regions where static recovery has taken place
Ask et al. Page 24 of 29
but without apparent grain boundary migration (recovery in connection with grain
boundary curvature on the other hand seems to have taken place to the right of
the region of interest, where the grain to the right has grown into the grain to the
left). There is some, but very little additional change between t = 1h and t = 2h,
and there is no definite indication for the moment that these small regions will
eventually expand.
The same behavior can be seen for the larger simulation geometry in figures 10 and
11. There are two grains a little bit above the center (below the spot marked with 1 in
figure 5, top right) where there is significant local reorientation and a similar pattern
of spots appear in the orientation field (figure 10 left) and corresponding regions of
apparent static recovery in the dislocation density (figure 11 right). In order to study
if this behavior is due to poor resolution of the localization taking place in the kink
bands, a second simulation was performed for the six grain geometry with a finer
mesh with 20449 (143 by 143) elements (the coarser mesh has 8464 elements). The
same type of behavior is again observed, as shown in figure 13. It is thus present
in two different geometries, and for the same geometry with two different finite
element sizes. Further investigation is necessary to determine the precise origin of
the irregularities in the orientation distribution, and if it is possible that they could
act as seeds for new grains and subsequent recrystallization. The evolution of the
other fields, notably the scalar dislocation density, can be explained by the time
evolution predicted by the model as a consequence of the heterogeneities in the
microstructure. The phase field φ always follows the orientation field, as indicated
in the previous examples, and this also activates the static recovery term in the
modified Kocks-Mecking-Teodosiu equations (50).
The second phenomena that may warrant further investigation appears in the
32 grain geometry, where one grain boundary exhibits signs of bulging. The grain
boundary can be located in figure 5 (top right) where it is near the spot marked
with a 2. This is shown in figures 14 and 15. The grain has a boundary near the
edge of the simulation box (in the periodic geometry), and so the change in the
profile to the very right in the figures is likely due to ordinary grain boundary
migration. However, a little to the left there is a small bulge that appears which
can not be explained by the geometry. This bulge is apparent in both the orientation
field in figure 14 and in the phase-field φ in figure 15 (top). In addition, it seems
to be associated with static recovery of the dislocation density ρ, shown in figure
15 (bottom). Again there is not much evolution however after the bulge initially
appears. It seems to stay at the same size between 1000 seconds and one hour in
figure 14. It is likely that it will be eaten up by the moving front to the right before
it evolves much further.
Summary and conclusionsMicrostructure evolution in deformed polycrystals is studied in this work by finite
element simulations, applying a constitutive model that combines Cosserat single
crystal plasticity with a phase-field model. It is shown that the model is capable of
taking into account significant microstructural changes. A major result is the impor-
tance of strain localization (slip bands) and curvature localization (kink bands) on
subsequent grain boundary formation and grain boundary migration. Simulations
Ask et al. Page 25 of 29
Figure 13 A zoom on a subgrain boundary generated in the six grain microstructure due to kinkband formation, c.f. figure 3 and 6 (the subgrain boundary appears in the lower left corner of themicrostructure in those figures). Two solutions are compared after six minutes of constantdeformation, which is when the very localized recovery of the dislocation density starts to appear.To the left, the same mesh was used as in the previous figure whereas in the images to the right,a finer mesh was used. From top to bottom, the images show the orientation, the dislocationdensity ρ and the norm of the curvature |∇θ|. The two solutions are not identical, but the samephenomena can be observed with localization near or inside regions with high local curvature anddislocation density.
with only one possible slip system are used to trigger the formation of kink bands
and obtain curvature localization. There is some clear evidence in the results of new
grain boundaries appearing – that is to say: it is demonstrated that the phase-field
variable is sensitive to regions of high dislocation density and that the coupled sim-
ulation model has inherited the quality of the KWC orientation phase-field model
to localize grain boundaries in regions of strong curvature.
Grain boundary migration is also present in the simulation results, both due to
stored energy (SSD densities) and to curvature. The most important driving pres-
sures for grain boundary migration are obtained when large stored energy gradients
occur between individual grains.
The presented simulations are still at their early stage (2D, small strain and
rotations) but they indicate that the model contains physical ingredients that are
responsible for subgrain boundary formation, formation of possible recrystallization
seeds, and evidence of bulging phenomenon. It will be necessary in a first step, which
is a current focus, to extend the model to perform more realistic 3D simulations. The
full 3D version of the model requires seven degrees of freedom at each material point,
rather than four in the 2D case. The increase in numerical cost is therefore very
significant. The main challenge, however, is the representation of the orientation
Ask et al. Page 26 of 29
Figure 14 A zoom on a grain boundary in the 32 grain microstructure, c.f. figure 5 and 10 (thesubgrain boundary appears in the lower right quadrant of the microstructure in those figures).From left to right, top to bottom, the orientation field around the grain boundary is shownimmediately after deformation, at 500, 1000 and 3600 seconds, respectively. The grain is on theedge of the simulation box, and so the change in the profile to the very right in the figure is likelydue to ordinary grain boundary migration. However, a little to the left there is a small bulge thatappears which can not be explained by the geometry.
Figure 15 A zoom on a grain boundary in the 32 grain microstructure, c.f. figure 5 and 10 (thesubgrain boundary appears in the lower right quadrant of the microstructure in those figures). Topline shows the phase-field φ and bottom line the dislocation density ρ. The images on the left areimmediately after the deformation is applied and the images on the right are after one hour ofconstant deformation. The maximum display value for the dislocation density has not been cappedin these images, unlike in previous figures.
field in 3D. In the plane case, the grain orientation is treated as a scalar field, with
Ask et al. Page 27 of 29
rotation only around one major axis. The fact that the orientation is restricted to
the plane in the 2D version of the model severely limits the types of slip systems that
can be studied, which will not be the case in a 3D formulation. However, if rotation
around several axes is considered, the interpolation of the variation of orientation
between two grains requires some consideration.
The large deformation formulation of the model has already been established [22]
and in a second step, its numerical implementation will allow to study processes
such as cold working and subsequent annealing but also to tackle the more complex
problem of dynamic recrystallization with concurrent deformation, nucleation and
grain growth. This would for example allow to investigate if the model can predict a
real threshold in stored energy for strain induced grain boundary migration to take
place. While the simulations performed in the present study showed some tendency
toward grain boundary bulging, it is not clear whether a critical work-hardening
can be an outcome of the model or if the model must be enhanced.
Author’s contributions
The coupled Cosserat phase-field theory has been developed by S. Forest, A. Ask and B. Appolaire. The
implementation of the simulation model has been done by K. Ammar and A. Ask together. All simulations in this
article were run by A. Ask. B. Appolaire contributed to the asymptotic analysis which was necessary to calibrate the
model parameters. The calibration was done by A. Ask. The paper has been written primarily by A. Ask and S.
Forest with important discussions and contributions from B. Appolaire.
Acknowledgements
Not applicable.
Availability of data and materials
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Funding
This project has received partial funding from the European Research Council (ERC) under the European Union’s
Horizon 2020 research and innovation program (grant agreement n 707392 MIGRATE).
Author details1Departement Materiaux et Structures, Onera, BP 72, 92322 Chatillon CEDEX, France. 2MINES ParisTech, PSL
University, Centre des materiaux (CMAT),CNRS UMR 7633, BP 87, 91003 Evry, France. 3Institut Jean Lamour,
CNRS-Universite de Lorraine, 2 allee Andre Guinier, 54000 Nancy, France.
References1. Asaro, R.J., Rice, J.R.: Strain localization in ductile single crystals. Journal of the Mechanics and Physics of
Solids 25(5), 309–338 (1977). doi:10.1016/0022-5096(77)90001-1
2. Forest, S.: Modeling slip, kink and shear banding in classical and generalized single crystal plasticity. Acta
Materialia 46(9), 3265–3281 (1998). doi:10.1016/S1359-6454(98)00012-3
3. Marano, A., Gelebart, L., Forest, S.: Intragranular localization induced by softening crystal plasticity: Analysis
of slip and kink bands localization modes from high resolution FFT-simulations results. Acta Materialia 15,
262–275 (2019). doi:10.1016/j.actamat.2019.06.010
4. Sedlacek, R., Blum, W., Kratochvil, J., Forest, S.: Subgrain formation during deformation : physical origin and
consequences. Metallurgical and Materials Transactions 33A, 319–327 (2002)
5. Ashby, M.F.: The deformation of plastically non–homogeneous materials. Philosophical Magazine 21, 399–424
(1970)
6. Bever, M.B., Holt, D.L., Titchener, A.L.: The stored energy of cold work. Progress in Materials Science 17,
5–177 (1973). doi:10.1016/0079-6425(73)90001-7
7. Humphreys, F.J., Hatherly, M.: Recrystallization and Related Annealing Phenomena. Elsevier, ??? (2004)
8. Gottstein, G., Shvindlerman, L.S.: Grain Boundary Migration in Metals. CRC Press, Berlin, ??? (2010)
9. Renk, O., Hohenwarter, A., Wurster, S., Pippan, R.: Direct evidence for grain boundary motion as the
dominant restoration mechanism in the steady-state regime of extremely cold-rolled copper. Acta Materialia 77,
401–410 (2014). doi:10.1016/j.actamat.2014.06.010
10. Taupin, V., Capolungo, L., Fressengeas, C., Das, A., Upadhyay, M.: Grain boundary modeling using an
elasto-plastic theory of dislocation and disclination. Journal of the Mechanics and Physics of Solids 61,
370–384 (2013)
11. Berbenni, S., Paliwal, B., Cherkaoui, M.: A micromechanics–based model for shear–coupled grain boundary
migration in bicrystals. International Journal of Plasticity 44, 68–94 (2013)
Ask et al. Page 28 of 29
12. Raabe, D., Becker, R.C.: Coupling of a crystal plasticity finite-element model with a probabilistic cellular
automaton for simulating primary static recrystallization in aluminium. Modelling and Simulation in Materials
Science and Engineering 8(4), 445 (2000)
13. Popova, E., Staraselski, Y., Brahme, A., Mishra, R.K., Inal, K.: Coupled crystal plasticity – probabilistic cellular
automata approach to model dynamic recrystallization in magnesium alloys. International Journal of Plasticity
66, 85–102 (2015). doi:10.1016/j.ijplas.2014.04.008. Plasticity of Textured Polycrystals In Honor of Prof. Paul
Van Houtte
14. Hallberg, H.: A modified level set approach to 2d modeling of dynamic recrystallization. Modelling and
Simulation in Materials Science and Engineering 21(8), 085012 (2013). doi:10.1088/0965-0393/21/8/085012
15. Blesgen, T.: A variational model for dynamic recrystallization based on Cosserat plasticity. Composites Part B:
Engineering 115, 236–243 (2017)
16. Takaki, T., Yamanaka, A., Higa, Y., Tomita, Y.: Phase-field model during static recrystallization based on
crystal-plasticity theory. Journal of Computer-Aided Materials Design 14(1), 75–84 (2007)
17. Takaki, T., Hisakuni, Y., Hirouchi, T., Yamanaka, A., Tomita, Y.: Multi-phase-field simulations for dynamic
recrystallization. Computational Materials Science 45(4), 881–888 (2009)
18. Guvenc, O., Henke, T., Laschet, G., Bottger, B., Apel, M., Bambach, M., Hirt, G.: Modelling of static
recrystallization kinetics by coupling crystal plasticity FEM and multiphase field calculations. Computer
Methods in Materials Science 13(2), 368–374 (2013)
19. Guvenc, O., Bambach, M., Hirt, G.: Coupling of crystal plasticity finite element and phase field methods for the
prediction of SRX kinetics after hot working. steel research international 85(6), 999–1009 (2014)
20. Vondrous, A., Bienger, P., Schreijag, S., Selzer, M., Schneider, D., Nestler, B., Helm, D., Monig, R.: Combined
crystal plasticity and phase-field method for recrystallization in a process chain of sheet metal production.
Computational Mechanics 55(2), 439–452 (2015)
21. Ask, A., Forest, S., Appolaire, B., Ammar, K.: Cosserat crystal plasticity with dislocation-driven grain boundary
migration. Journal of Micromechanics and Molecular Physics 03(03n04), 1840009 (2018).
doi:10.1142/S242491301840009X
22. Ask, A., Forest, S., Appolaire, B., Ammar, K.: A cosserat–phase field theory of crystal plasticity and grain
boundary migration at finite deformation. Continuum Mechanics and Thermodynamics 31(4), 1109–1141
(2018). doi:10.1007/s00161-018-0727-6
23. Admal, N.C., Po, G., Marian, J.: A unified framework for polycrystal plasticity with grain boundary evolution.
International Journal of Plasticity (2018). doi:10.1016/j.ijplas.2018.01.014
24. Kobayashi, R., Warren, J.A., Carter, W.C.: A continuum model of grain boundaries. Physica D 140(1-2),
141–150 (2000)
25. Warren, J.A., Kobayashi, R., Lobkovsky, A.E., Carter, W.C.: Extending phase field models of solidification to
polycrystalline materials. Acta Materialia 51(20), 6035–6058 (2003)
26. Abrivard, G., Busso, E.P., Forest, S., Appolaire, B.: Phase field modelling of grain boundary motion driven by
curvature and stored energy gradients. Part I: theory and numerical implementation. Philosophical Magazine
92, 3618–3642 (2012)
27. Abrivard, G., Busso, E.P., Forest, S., Appolaire, B.: Phase field modelling of grain boundary motion driven by
curvature and stored energy gradients. Part II: Application to recrystallisation. Philosophical Magazine 92,
3643–3664 (2012)
28. Ask, A., Forest, S., Appolaire, B., Ammar, K., Salman, O.U.: A cosserat crystal plasticity and phase field theory
for grain boundary migration. Journal of the Mechanics and Physics of Solids 115, 167–194 (2018).
doi:10.1016/j.jmps.2018.03.006
29. Eringen, A.C., Kafadar, C.B.: Part I. Polar Field Theories. In: Eringen, A.C. (ed.) Continuum Physics, pp. 1–73.
Academic Press, ??? (1976)
30. Forest, S., Cailletaud, G., Sievert, R.: A Cosserat theory for elastoviscoplastic single crystals at finite
deformation. Archives of Mechanics 49(4), 705–736 (1997)
31. Nye, J.F.: Some geometrical relations in dislocated crystals. Acta Metall. 1, 153–162 (1953)
32. Forest, S., Sievert, R.: Elastoviscoplastic constitutive frameworks for generalized continua. Acta Mechanica
160, 71–111 (2003)
33. Gurtin, M.E.: Generalized Ginzburg-Landau and Cahn-Hilliard equations based on a microforce balance. Physica
D: Nonlinear Phenomena 92(3), 178–192 (1996)
34. Gurtin, M.E., Lusk, M.T.: Sharp-interface and phase-field theories of recrystallization in the plane. Physica D:
Nonlinear Phenomena 130(1–2), 133–154 (1999)
35. Gurtin, M.E.: A gradient theory of single-crystal viscoplasticity that accounts for geometrically necessary
dislocations. Journal of the Mechanics and Physics of Solids 50(1), 5–32 (2002)
36. Kobayashi, R., Giga, Y.: Equations with Singular Diffusivity. Journal of Statistical Physics 95(5-6), 1189–1220
(1999)
37. Ohno, N., Okumura, D.: Higher–order stress and grain size effects due to self–energy of geometrically necessary
dislocations. Journal of the Mechanics and Physics of Solids 55, 1879–1898 (2007)
38. Mesarovic, S.D., Forest, S., Jaric, J.P.: Size-dependent energy in crystal plasticity and continuum dislocation
models. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 471(2175),
20140868–19 (2015)
39. Wulfinghoff, S., Forest, S., Bohlke, T.: Strain gradient plasticity modeling of the cyclic behavior of laminate
microstructures. Journal of the Mechanics and Physics of Solids 79, 1–20 (2015).
doi:10.1016/j.jmps.2015.02.008
40. Blesgen, T.: Deformation patterning in three-dimensional large-strain Cosserat plasticity. Mechanics Research
Communications 62, 37–43 (2014)
41. Forest, S., Barbe, F., Cailletaud, G.: Cosserat modelling of size effects in the mechanical behaviour of
polycrystals and multiphase materials. International Journal of Solids and Structures 37, 7105–7126 (2000)
42. Mayeur, J.R., McDowell, D.L., Bammann, D.J.: Dislocation–based micropolar single crystal plasticity:
Ask et al. Page 29 of 29
Comparison of multi– and single criterion theories. Journal of the Mechanics and Physics of Solids 59, 398–422
(2011)
43. Mayeur, J.R., McDowell, D.L.: A comparison of gurtin type and micropolar theories of generalized single crystal
plasticity. International Journal of Plasticity 57, 29–51 (2014). doi:10.1016/j.ijplas.2014.01.010
44. Sedlacek, R., Forest, S.: Non-local plasticity at microscale : A dislocation-based model and a Cosserat model.
physica status solidi (b) 221, 583–596 (2000)
45. Forest, S., Sedlacek, R.: Plastic slip distribution in two–phase laminate microstructures: Dislocation–based vs.
generalized–continuum approaches. Philosophical Magazine A 83, 245–276 (2003)
46. Forest, S.: Some links between Cosserat, strain gradient crystal plasticity and the statistical theory of
dislocations. Philosophical Magazine 88, 3549–3563 (2008)
47. Gerard, C., Bacroix, B., Bornert, M., Cailletaud, G., Crepin, J., Leclercq, S.: Hardening description for fcc
materials under complex loading paths. Computational Materials Science 45(3), 751–755 (2009).
doi:10.1016/j.commatsci.2008.08.027. Proceedings of the 17th International Workshop on Computational
Mechanics of Materials
48. Cheong, K.-S., Busso, E.P.: Discrete dislocation density modelling of single phase fcc polycrystal aggregates.
Acta Materialia 52(19), 5665–5675 (2004). doi:10.1016/j.actamat.2004.08.044
49. Lobkovsky, A.E., Warren, J.A.: Sharp interface limit of a phase-field model of crystal grains. Physical Review E
63(5), 051605 (2001)
50. Wolf, D.: Structure-energy correlation for grain boundaries in f.c.c. metals—iii. symmetrical tilt boundaries.
Acta Metallurgica et Materialia 38(5), 781–790 (1990). doi:10.1016/0956-7151(90)90030-K
51. litold Missol: Orientation dependence of grain-boundary energy in metals in the view of a pseudoheterophase
dislocation core model. Acta Metallurgica 24(1), 73–79 (1976). doi:10.1016/0001-6160(76)90149-8
52. Tschopp, M.A., Coleman, S.P., McDowell, D.L.: Symmetric and asymmetric tilt grain boundary structure and
energy in cu and al (and transferability to other fcc metals). Integrating Materials and Manufacturing
Innovation 4(1), 176–189 (2015). doi:10.1186/s40192-015-0040-1
53. Z–set package: Non-linear material & structure analysis suite, www.zset-software.com (2013)
54. Vandermeer, R.A., Jensen, D.J., Woldt, E.: Grain boundary mobility during recrystallization of copper.
Metallurgical and Materials Transactions A 28(3), 749–754 (1997). doi:10.1007/s11661-997-0061-2
55. Quey, R., Dawson, P.R., Barbe, F.: Large-scale 3d random polycrystals for the finite element method:
Generation, meshing and remeshing. Computer Methods in Applied Mechanics and Engineering 200(17),
1729–1745 (2011). doi:10.1016/j.cma.2011.01.002